between R&D and productivity. Workshop participants suggested several steps to overcome these barriers, ranging from integration of databases across agencies to the inclusion of a single item of identifying information in a standard survey form. For example, Adam Jaffe suggested that companies be asked on the RD-1 form to list the names of the entities under which they hold patents, allowing researchers to match the R&D data to patent data by industry.
The workshop was not designed to assign priorities to these suggestions nor to examine their cost and feasibility. Nevertheless, Steve Landefeld made the encouraging observation that although considerable economic data are acquired by BEA under the condition that they not be linked to other data, a 1990 law allows BEA's foreign direct investment and services data to be linked to Census Bureau establishment data.7 Furthermore, proposed legislation would allow broad data sharing among BEA, the Census Bureau, the Bureau of Labor Statistics, and most of the principal U.S. statistical agencies.
High-technology and service-sector industries have not been adequately represented in the national technology and innovation-related databases. Many workshop participants endorsed NSF's current efforts to expand the coverage to service and emerging high-technology industries but agreed that a good deal of fundamental research on the structure and nature of industrial innovation systems in these industries is needed to inform new data collection efforts. Unfortunately, the Standard Industrial Classification system, in spite of its recent revision, still does not represent important new areas of economic activity such as biotechnology.
Many workshop participants endorsed the proposition that collection and reporting of R&D and innovation survey information at the level of the business unit rather than the firm level or establishment level would not only enhance its utility but also improve its quality. Survey instruments should be addressed to the most appropriate person or unit in the firm. In large multiproduct firms, R&D units may not have accurate information on firm-level sales, and specific units often report operational figures that do not match those used in other units or by headquarters management.
The NSF collects information on R&D activities at the state level, but the state is not the most useful level of geographic aggregation for analyzing innova-